ASAPPlanner turns SQL, PromQL, and MetricsQL query workloads into legal candidate plans that may use Approximate Streaming Analytics Primitives (ASAPs), such as sketches and exact summaries. It normalizes language-specific queries into a shared representation, enumerates semantically equivalent alternatives, checks each against the queries' accuracy and latency requirements, and selects a physical plan under the deployment's cost model. Downstream systems deploy and execute that plan.
- New to the repository? Read the planner pipeline, then the glossary.
- Want to run a query? Follow Run and inspect a query.
- Want to see how a workload is planned, stage by stage? Use the Stage Viewer.
- Embedding Planner? Use the library API guide.
- Extending Planner? Read the planning stages design, then the ASAP-aware mapping architecture, which maps each stage to its code, then extend ASAP-aware mapping.
- Evaluating a design? Browse the design documentation, developer documentation, and user guides.
The documentation map gives each audience a complete reading path.
cargo build
cargo test --workspace
No external setup is required for the standard build and test suite.
Each Git worktree has its own target/, so a new worktree rebuilds every
dependency (DataFusion, Arrow, HiGHS, ...). kache
caches compiler outputs on your machine and restores them in other worktrees:
cargo install kache --locked
kache init
kache init makes kache Cargo's compiler wrapper and starts its background
daemon. Its other offers (shell shims, a login service) are optional. To use
kache without kache init, set rustc-wrapper = "kache" under [build] in
~/.cargo/config.toml.
kache stats --last-buildshows the hit rate;kache explain <crate>says why a crate missed.KACHE_DISABLED=1 cargo buildbypasses the cache for one build.- The HiGHS C++ build (
highs-sys) still reruns in each new worktree, because its output records its absolute build path. - The cache needs about 14 GiB per toolchain and profile.